Abstarct: Voice over LTE (VoLTE) technology, as the dominant standard for voice communications in modern mobile networks, has brought significant advantages. However, the transition to an all-IP architecture has introduced new attack surfaces, particularly concerning location privacy. Previous research has primarily focused on information leakage through explicit message content. This thesis introduces a novel, stealthy, and low-cost class of attack that exploits a timing side-channel in the VoLTE call setup process to infer a user's location. The core hypothesis of this research is that minor, millisecond-level variations in the timing of Session Initiation Protocol (SIP) message exchanges create a unique and location-dependent temporal fingerprint. To test this hypothesis, a comprehensive frxamework was designed and implemented. Within this frxamework, over 50,000 call logs were collected via the Android Debug Bridge (ADB) tool by making repeated, automated calls to target devices in 9 different cities nationwide and 6 distinct regions within a single city. Subsequently, by extracting precise temporal features from the SIP message sequences, these features were used as input to train five different machine learning models, including Random Forest and Recurrent Neural Networks. The experimental results successfully validated the research hypothesis. The Random Forest model achieved an accuracy of 85.3% in the inter-city classification scenario, while the Gradient Boosting model reached 79.4% accuracy in the intra-city scenario. Feature importance analysis revealed that the models intelligently adapt their decision-making criteria baxsed on the geographical scale of the problem. This research demonstrates the existence of a fundamental and practical vulnerability in VoLTE networks, rooted in the physical and topological characteristics of the network, against which traditional countermeasures are ineffective.